The Ghost in the Machine: Why Engaging AI Personalities Are Becoming B2B Playgrounds

A company generating 100 million user messages in three months and used by hundreds of thousands of people sounds like a runaway success story, yet Poke, the AI text assistant, was struggling to turn a profit. Its acquisition by Cognition, the AI coding startup behind Devin, for a “low nine figures” valuation, isn’t just about integrating a chatty interface into a developer tool. It’s a stark reminder that even compelling consumer AI, especially conversational agents, faces immense financial pressure unless it can directly justify its operational costs. This deal lays bare a structural implication: the most engaging, human-like AI personalities, once hailed as the future of consumer tech, are being repurposed as features in more lucrative enterprise or niche B2B products.

The Unspoken Cost of Being a ‘Friend’

Poke’s unique selling proposition, as co-founder Marvin von Hagen explained, was its ability to chat like a friend, incorporating slang and humor, rather than functioning as a mere utility. This model led to a significant user engagement metric, exceeding 100 million messages exchanged in just three months. Yet, von Hagen candidly admitted Poke was “expensive to run” and “difficult to turn a profit.” This isn’t a minor detail; it’s the quiet truth behind many high-burn, high-interaction consumer AI plays. The computational overhead for maintaining sophisticated conversational models, especially across multiple messaging platforms like iMessage, SMS, and WhatsApp, is substantial. Free or low-cost user access, common in early-stage consumer AI, rarely covers these soaring infrastructure expenses.

This transaction reveals a brutal incentive: even if you build an agent “people love,” as Cognition’s Scott Wu noted, that love doesn’t automatically translate into sustainable revenue. The path to profitability for general-purpose AI companions remains elusive. Unlike traditional software, where marginal costs often diminish with scale, each additional AI interaction carries a non-trivial computational cost, turning user delight into a balance sheet liability without a clear conversion funnel.

From Consumer Delight to Developer Utility

Cognition’s primary interest in Poke isn’t to launch a new consumer chatbot. It’s to graft Poke’s personable interaction model onto Devin, its AI coding assistant. The vision is clear: make Devin feel less like a rigid tool and more like a “persistent co-worker,” as von Hagen put it, one who can “make a joke” and remember tasks across sessions. This immediately shifts the economic calculus. In a developer workflow, personality isn’t just a pleasant add-on; it’s a potential productivity enhancer, reducing friction and increasing adoption for a tool that businesses pay for. The value proposition moves from abstract “friendship” to concrete “workflow efficiency” and “developer experience,” metrics that directly impact enterprise bottom lines.

This pivot is a smart survival strategy for advanced conversational AI. Instead of competing in a crowded, low-margin consumer market against the likes of ChatGPT or even bespoke app-integrated assistants, Poke’s capabilities find a new, defensible niche. Its approval for Apple’s Messages for Business platform in June hinted at this shift, acknowledging that a structured, business-oriented interaction model might unlock value where open-ended consumer chat could not. The integration with Cognition’s SWE-1.7 model for developer tasks solidifies this trajectory, moving beyond mere banter to orchestrate complex coding pull requests.

The Looming Consolidation of AI Interfaces

What we’re witnessing is a subtle but profound consolidation. The highly personalized, engaging interfaces that once defined cutting-edge consumer AI are being absorbed into more specialized applications. This isn’t just about an acquisition; it’s a blueprint for how novel AI interaction paradigms will find their footing. It suggests that the front-end personality layer, arguably the most human-facing aspect of AI, will increasingly become a feature of deeper, more functional platforms rather than a standalone product. The Silicon Valley narrative often prioritizes scaling to billions of users, but this story implies that for sophisticated AI, the path to value might actually lie in narrower, high-value contexts.

This trend is not isolated. We’ve seen similar patterns in other adjacent technologies, such as voice assistants struggling for independent monetization and subsequently integrated into smart home ecosystems or automotive interfaces. The market for general-purpose, open-ended conversational AI is proving incredibly difficult to sustain as a primary business model, especially as powerful foundation models from OpenAI, Anthropic, or Google make basic conversational capabilities increasingly commoditized. The truly valuable differentiation isn’t just what an AI can say, but how it says it, and crucially, who pays for that experience. For now, it seems the developers and enterprises hold that purse. The idea that people will pay a premium for a purely companionate AI, especially when core models are so powerful, is a pleasant fiction that’s rapidly dissipating.

Cognition’s move implies a future where the most endearing AI personalities are not our ubiquitous digital friends, but specialized agents tailored for specific, high-stakes tasks—like writing code. The ghost in the machine will no longer be free-floating; it will be firmly tethered to a purpose, and critically, a profit center.